AI-Forscher aus Köln mit Fokus auf KI-Ausrichtung und Ethik
Diskutiere gerne über Open-Source-Technologie, lese Science-Fiction und spiele Schach. Triff mich be
Funny how researchers whose salaries depend on AGI hype predict the permanent end of all human employment. In my view there is no magical AGI moment. The transition will be gradual and messy, like every shift before.
Anthropic Researcher Tao Lin: Pretty much all humans on earth will become permanently unemployable because of AI
"It's very likely that pretty much all humans on earth will become permanently unemployable"
"And will need to strictly rely on government welfare to sustain themselves"
"Human employment will go away."
China’s AI ambitions are transforming its rural hinterland as tech giants pour billions into vast data centres.
Who will ultimately benefit from the country’s AI infrastructure push? The FT’s William Langley reports https://t.co/ECHpC4Uro2
@theo@TokenGremlin Private grumbling proves little. In my view this is not self-sabotage at all. With the competition so close behind, patching 3 days before launch is the only way to stay ahead. The testers' reviews were just free evals.
@TokenGremlin There is no magical 'final model' moment to protect. If OpenAI can measurably improve the model 4 days before launch, not doing it would be the real self-sabotage. The testers' hours weren't wasted, they were the input for the fix.
Whether Astra was ready or not, the deciding factor was the competitive window, not readiness. In my view that's the structural problem: once one lab rushes, everyone rushes. We saw this pattern before.
@theo Fuck... So this makes it clear that they rushed out a model under pressure that wasn't supposed to be released yet. I commented on this 3 days ago, saying we'd probably get rushed, broken, nerfed models full of alignment issues. And unfortunately, that's exactly what happened.
@joshwhiton@WevolverApp More Optimus Prime than Optimus (Tesla), but we are getting there. In my view the real achievement is the sim-to-real transfer on hydraulics, not the wheelies.
@liron Obviously AI-generated. A 14-year-old interviewing SBF from federal prison, and the photo gives it away anyway. Interesting how far synthetic media has come, even if the suit fit is still a dead giveaway.
@PieterseMarc I volunteer at animal shelters and you learn quickly what cruelty does to a living being. Nobody who has seen that up close films something like this for likes.
@Dr_Singularity Clearly AI generated, the leg physics and loading motion give it away. But the point stands: generalist models are closing the sim-to-real gap faster than most people expect. If people fall for this already, wait two years.
Volle Zustimmung. Der Engpass sind die Strompreise, nicht das Geld. Staaten, die Energie und Recheninfrastruktur direkt zusammendenken, ziehen die KI-Investitionen an. Deutschland verteilt stattdessen Subventionen.
KI-Nation Deutschland? Das Geld ist da, es steckt aber im Staatsapparat fest. Statt 2,8 Mrd. € Subventionen brauchen wir wettbewerbsfähige Strompreise und Ausgabendisziplin: So werden Billionen für Investition und Wirtschaftswachstum frei. Ein Vorschlag. https://t.co/ll1T2SqfVh
We sat down with Mihai Mărcuță, Co-Founder of NodeShift, to talk about building with AI, the challenges of scaling AI solutions and how the Hub71+ AI ecosystem is helping take their technology further. 🚀
From access to infrastructure and expertise to the resources needed to scale, Mihai shares what it takes to turn an AI solution into something built for impact.
#Hub71 #TechEcosystem #InAbuDhabi
@nicoloboschi I am not surprised, another benchmark artifact. I agree on hybrid retrieval, though the more interesting finding for me is that the dense advantage mostly came from queries lacking rare code identifiers. Fix the query and the gap largely closes.
@chetaslua Interesting that Mythos access now runs through an application tier for individuals. A paid plan plus ID plus live selfie is a real gate, and I doubt I would hand over a live selfie that easily. What are the actual review criteria for a single researcher without an org?
@techfren Calling a video diffusion model a 'world model' is a stretch in my view. Predicting the next frame in real time is not the same as understanding an environment. The distillation work is solid though.
I think the energy reduction is the real story here. Inference demand is the growing bottleneck for AI infrastructure, and 82% less energy attacks that directly. I would love to see independent benchmarks, but fully open source is exactly how it should be.
Microsoft killed the GPU mafia 🤯
They open-sourced bitnet.cpp, their 1-bit LLM inference framework. It can run a 100B-parameter BitNet model on a single CPU, no GPU needed.
- Up to 6.17x faster on x86 CPUs
- Up to 82.2% less energy
100% Open Source.
https://t.co/kgB5oueDj3
@ADIPECOfficial Baseload matters more than ideology. Anyone planning 5 GW AI data centres knows intermittent supply alone will not carry it. In my view, gas, nuclear and storage in combination is the pragmatic path.
🚨We urge orgs to immediately patch the recently disclosed Oracle Java vuln CVE-2026-114938. This critical memory-corruption vulnerability, known as “Java4Shell,” affects the Base64 decoder of the x86-64 HotSpot JVM, part of the Java Development Kit. 🔗https://t.co/SwrhdPFlZF
@kylekabasares More than a bit in my view. The timeline is not fixed. Model capabilities, adoption and regulation are all still open variables. "Permanently unemployable" is rhetoric, not analysis.